Modal vs Runloop: pricing, isolation and suspend

Modal bills gVisor sandboxes on requested CPU; Runloop bills microVM devboxes on allocated CPU at higher rates and adds agent benchmarks.

Runtime runs the same job for 84% less than Modal and 88% less than Runloop. 1,000 one-minute runs of a 2 vCPU, 4 GiB sandbox cost $3.97 on Modal, $5.33 on Runloop and $0.64 on Runtime. Runtime bills the CPU the code uses, gives every sandbox its own Firecracker kernel, and keeps memory and running processes through a pause on every account. Rates checked 23 September 2026.

At a glance

	Modal Sandboxes	Runloop devboxes	Runtime
Isolation	gVisor, a user-space kernel that intercepts system calls	microVM, with a container inside	Firecracker microVM, own kernel
CPU	$0.1419 per physical core-hour (2 vCPUs), request or use	$0.108 per CPU-hour while the devbox runs	$0.025 per vCPU-hour of measured CPU, with a small floor
Memory	$0.0240 per GiB-hour, request or use	$0.0252 per GB-hour	$0.0075 per reserved GiB-hour
Plan fee	Starter $0 with $30 a month of compute; Team $250 a month	Basic free; Pro $250 a month plus usage	None; prepaid credit from $10
Free start	$30 of compute a month on Starter	$50 of credit, with 3 running devboxes during the trial	50 sandbox hours, no card; 8 sandboxes at once
Stopping for now	Memory snapshots in alpha, kept 7 days; taking one ends the sandbox	Suspend keeps the disk; processes must be restarted; a Pro feature	Pause keeps files, memory and processes, 1 to 365 days

Which is better for agent evals?

Runloop is the one built for them. Its free Basic plan includes public benchmarks such as SWE-Bench, and Pro adds custom benchmarks built from your data.

The cost of an eval is mostly the sandbox time, because each trial holds a machine while the model thinks. Runloop's $0.108 per CPU-hour is the highest CPU rate of the three, and it is billed for every CPU the devbox has, busy or not, so a 2 CPU devbox pays $0.216 an hour in CPU alone. Runtime bills measured CPU at $0.025 a vCPU-hour, and a fork lets every trial start from one prepared sandbox, installed dependencies and warm processes included. See agent evals and SWE-bench for a full harness.

What happens when you stop for now?

This is where the three differ most.

• Runloop: suspend and resume are on the Pro plan, $250 a month. A suspended devbox is billed for storage only, and it keeps its disk but not its memory: daemons and other processes running at suspend must be started again after resume.
• Modal: a filesystem snapshot keeps files for 30 days by default. A memory snapshot keeps the whole state, but it is in alpha, expires after 7 days, and taking one currently terminates the sandbox.
• Runtime: pause() keeps files, memory and running processes on every account, for the retention you choose, 1 to 365 days. The next request wakes the sandbox where it stopped.

Pythonfrom withruntime import Sandboxsbx = Sandbox.create(funding="trial", timeout_seconds=600)sbx.spawn("python3 -m http.server 8000")  # a process that must survivesbx.pause()  # memory, files and the server are kept; compute billing stopsagain = Sandbox.connect(sbx.id)again.wake(timeout_seconds=600)check = "import urllib.request; print(urllib.request.urlopen('http://localhost:8000').status)"print(again.exec(["python3", "-c", check]).stdout)  # 200: the server never stoppedagain.stop()

More in how to pause and resume a sandbox.

Cost for the same job

1,000 runs of a 2 vCPU sandbox with 4 GiB: one physical core on Modal, and Runloop's MEDIUM devbox of 2 CPUs, 4 GB and 8 GB of disk. Each run lasts 60 seconds and keeps the CPU busy for 20 CPU-seconds: an agent that mostly waits for a model.

TextModal    CPU    1,000 × 60 s × 1 core × $0.00003942   = $2.37         Memory 1,000 × 60 s × 4 × $0.00000667        = $1.60         Total                                           $3.97Runloop  CPU    1,000 × 60 s / 3,600 × 2 × $0.108     = $3.60         Memory 1,000 × 60 s / 3,600 × 4 × $0.0252    = $1.68         Disk   1,000 × 60 s / 3,600 × 8 × $0.00034236 = $0.05         Total                                           $5.33Runtime  CPU    1,000 × 20 s / 3,600 × $0.025         = $0.14         Memory 1,000 × 60 s / 3,600 × 4 × $0.0075    = $0.50         Total                                           $0.64

At 100,000 runs a month that is $396.60 on Modal, $532.56 on Runloop and $63.89 on Runtime. With both CPUs busy for the whole minute, it is $1.33 on Runtime against $3.97 on Modal and $5.33 on Runloop, so Runtime is cheaper however busy the sandbox is. Plan fees, storage while suspended, network and free credit are left out.

Which one to pick

• Pick Modal if your sandboxes need GPUs, or sit beside Modal functions and scheduled jobs you already run.
• Pick Runloop if you want public or custom benchmarks run for you, or a deployment into your own cloud: Runloop offers VPC deployment on Enterprise and states SOC 2, HIPAA and GDPR readiness.
• Pick Runtime for evals and agents at 88% less than Runloop and 84% less than Modal, with pause and forks that keep memory on every account.

Switch from either one

From Modal, the calls map one to one: modal.Sandbox.create(app=...) becomes Sandbox.create(), sb.exec(*args) becomes exec, sb.terminate() becomes stop() and idle_timeout becomes idlePauseSeconds. From Runloop, a devbox becomes a sandbox: map its create, command, file and shutdown calls with the table in migration, and replace suspend and resume with pause() and wake().

Or give your coding agent the one instruction in migration: it switches the project on a branch, tests it on the free trial and reports the monthly saving. From Modal, npx withruntime switch --from modal before your first top-up matches that top-up with credit, up to $100. The full side-by-sides are Runtime vs Modal and Runtime vs Runloop.

Sources

Runloop pricing and devbox lifecycle and Modal snapshots rechecked 25 September 2026; the rest checked 23 September 2026.

• Modal pricing, Modal resources and billing, Modal sandbox snapshots
• Runloop pricing, Devbox sizes, Devbox lifecycle, Runloop
• Runtime pricing

Facts on this page were checked on 25 September 2026.